Type in the airport codes for your departure and arrival cities and the date and site come back with predictions about the probability of cancellation and delay for different airlines serving the route. For example, here are the results for a trip from Chicago (ord) to Hartford (bdl):
And in true Super Crunching fashion, DelayCast not only predicts but gives the 90 percent confidence interval for the prediction. The site’s F.A.Q. explains why their predictions are better than the government’s on-time statistics.
Now what I’d like to see is a DelayCast for other parts of my life.
In particular, I’d like to have a statistical algorithm for accurately “translating” the delay predictions of pilots. My strong sense is that when a pilot comes on the intercom and tells you that the plane will be taking off in 10 minutes, the passengers should expect the plane to leave in 15 or 20 minutes. (The true distribution is skewed right and a good Bayesian would also increase the probability that the flight would be canceled).
Instead of trying to get pilots to speak more honestly about delays, I think we’re more likely to be able to produce a translation program:
When pilot says X, passengers should hear Y.
You can help create such a program. For the next week if you’re traveling by air, please pay attention to all the predicted delays and compare them to the actual delays — and post your numbers as a comment to this post. (I’m a bit scared that people will disproportionately report when they are frustrated by an unexpectedly long delay — so the data is likely to be unrepresentative.)
But if we get enough data, I’ll crunch some numbers. The exercise underscores the needless (or maybe I should say self-interested) bias in many human delay predictions.
When a home contractor says your kitchen renovation will be done in two weeks, you’d be foolish to start sending dinner invitations. But it would be nice to know whether two weeks means three weeks or two months.